1/*
   2 * Copyright (C) 2001 Sistina Software (UK) Limited.
   3 * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved.
   4 *
   5 * This file is released under the GPL.
   6 */
   7
   8#include "dm.h"
   9
  10#include <linux/module.h>
  11#include <linux/vmalloc.h>
  12#include <linux/blkdev.h>
  13#include <linux/namei.h>
  14#include <linux/ctype.h>
  15#include <linux/string.h>
  16#include <linux/slab.h>
  17#include <linux/interrupt.h>
  18#include <linux/mutex.h>
  19#include <linux/delay.h>
  20#include <linux/atomic.h>
  21
  22#define DM_MSG_PREFIX "table"
  23
  24#define MAX_DEPTH 16
  25#define NODE_SIZE L1_CACHE_BYTES
  26#define KEYS_PER_NODE (NODE_SIZE / sizeof(sector_t))
  27#define CHILDREN_PER_NODE (KEYS_PER_NODE + 1)
  28
  29/*
  30 * The table has always exactly one reference from either mapped_device->map
  31 * or hash_cell->new_map. This reference is not counted in table->holders.
  32 * A pair of dm_create_table/dm_destroy_table functions is used for table
  33 * creation/destruction.
  34 *
  35 * Temporary references from the other code increase table->holders. A pair
  36 * of dm_table_get/dm_table_put functions is used to manipulate it.
  37 *
  38 * When the table is about to be destroyed, we wait for table->holders to
  39 * drop to zero.
  40 */
  41
  42struct dm_table {
  43	struct mapped_device *md;
  44	atomic_t holders;
  45	unsigned type;
  46
  47	/* btree table */
  48	unsigned int depth;
  49	unsigned int counts[MAX_DEPTH];	/* in nodes */
  50	sector_t *index[MAX_DEPTH];
  51
  52	unsigned int num_targets;
  53	unsigned int num_allocated;
  54	sector_t *highs;
  55	struct dm_target *targets;
  56
  57	struct target_type *immutable_target_type;
  58	unsigned integrity_supported:1;
  59	unsigned singleton:1;
  60
  61	/*
  62	 * Indicates the rw permissions for the new logical
  63	 * device.  This should be a combination of FMODE_READ
  64	 * and FMODE_WRITE.
  65	 */
  66	fmode_t mode;
  67
  68	/* a list of devices used by this table */
  69	struct list_head devices;
  70
  71	/* events get handed up using this callback */
  72	void (*event_fn)(void *);
  73	void *event_context;
  74
  75	struct dm_md_mempools *mempools;
  76
  77	struct list_head target_callbacks;
  78};
  79
  80/*
  81 * Similar to ceiling(log_size(n))
  82 */
  83static unsigned int int_log(unsigned int n, unsigned int base)
  84{
  85	int result = 0;
  86
  87	while (n > 1) {
  88		n = dm_div_up(n, base);
  89		result++;
  90	}
  91
  92	return result;
  93}
  94
  95/*
  96 * Calculate the index of the child node of the n'th node k'th key.
  97 */
  98static inline unsigned int get_child(unsigned int n, unsigned int k)
  99{
 100	return (n * CHILDREN_PER_NODE) + k;
 101}
 102
 103/*
 104 * Return the n'th node of level l from table t.
 105 */
 106static inline sector_t *get_node(struct dm_table *t,
 107				 unsigned int l, unsigned int n)
 108{
 109	return t->index[l] + (n * KEYS_PER_NODE);
 110}
 111
 112/*
 113 * Return the highest key that you could lookup from the n'th
 114 * node on level l of the btree.
 115 */
 116static sector_t high(struct dm_table *t, unsigned int l, unsigned int n)
 117{
 118	for (; l < t->depth - 1; l++)
 119		n = get_child(n, CHILDREN_PER_NODE - 1);
 120
 121	if (n >= t->counts[l])
 122		return (sector_t) - 1;
 123
 124	return get_node(t, l, n)[KEYS_PER_NODE - 1];
 125}
 126
 127/*
 128 * Fills in a level of the btree based on the highs of the level
 129 * below it.
 130 */
 131static int setup_btree_index(unsigned int l, struct dm_table *t)
 132{
 133	unsigned int n, k;
 134	sector_t *node;
 135
 136	for (n = 0U; n < t->counts[l]; n++) {
 137		node = get_node(t, l, n);
 138
 139		for (k = 0U; k < KEYS_PER_NODE; k++)
 140			node[k] = high(t, l + 1, get_child(n, k));
 141	}
 142
 143	return 0;
 144}
 145
 146void *dm_vcalloc(unsigned long nmemb, unsigned long elem_size)
 147{
 148	unsigned long size;
 149	void *addr;
 150
 151	/*
 152	 * Check that we're not going to overflow.
 153	 */
 154	if (nmemb > (ULONG_MAX / elem_size))
 155		return NULL;
 156
 157	size = nmemb * elem_size;
 158	addr = vzalloc(size);
 159
 160	return addr;
 161}
 162EXPORT_SYMBOL(dm_vcalloc);
 163
 164/*
 165 * highs, and targets are managed as dynamic arrays during a
 166 * table load.
 167 */
 168static int alloc_targets(struct dm_table *t, unsigned int num)
 169{
 170	sector_t *n_highs;
 171	struct dm_target *n_targets;
 172	int n = t->num_targets;
 173
 174	/*
 175	 * Allocate both the target array and offset array at once.
 176	 * Append an empty entry to catch sectors beyond the end of
 177	 * the device.
 178	 */
 179	n_highs = (sector_t *) dm_vcalloc(num + 1, sizeof(struct dm_target) +
 180					  sizeof(sector_t));
 181	if (!n_highs)
 182		return -ENOMEM;
 183
 184	n_targets = (struct dm_target *) (n_highs + num);
 185
 186	if (n) {
 187		memcpy(n_highs, t->highs, sizeof(*n_highs) * n);
 188		memcpy(n_targets, t->targets, sizeof(*n_targets) * n);
 189	}
 190
 191	memset(n_highs + n, -1, sizeof(*n_highs) * (num - n));
 192	vfree(t->highs);
 193
 194	t->num_allocated = num;
 195	t->highs = n_highs;
 196	t->targets = n_targets;
 197
 198	return 0;
 199}
 200
 201int dm_table_create(struct dm_table **result, fmode_t mode,
 202		    unsigned num_targets, struct mapped_device *md)
 203{
 204	struct dm_table *t = kzalloc(sizeof(*t), GFP_KERNEL);
 205
 206	if (!t)
 207		return -ENOMEM;
 208
 209	INIT_LIST_HEAD(&t->devices);
 210	INIT_LIST_HEAD(&t->target_callbacks);
 211	atomic_set(&t->holders, 0);
 212
 213	if (!num_targets)
 214		num_targets = KEYS_PER_NODE;
 215
 216	num_targets = dm_round_up(num_targets, KEYS_PER_NODE);
 217
 218	if (alloc_targets(t, num_targets)) {
 219		kfree(t);
 220		t = NULL;
 221		return -ENOMEM;
 222	}
 223
 224	t->mode = mode;
 225	t->md = md;
 226	*result = t;
 227	return 0;
 228}
 229
 230static void free_devices(struct list_head *devices)
 231{
 232	struct list_head *tmp, *next;
 233
 234	list_for_each_safe(tmp, next, devices) {
 235		struct dm_dev_internal *dd =
 236		    list_entry(tmp, struct dm_dev_internal, list);
 237		DMWARN("dm_table_destroy: dm_put_device call missing for %s",
 238		       dd->dm_dev.name);
 239		kfree(dd);
 240	}
 241}
 242
 243void dm_table_destroy(struct dm_table *t)
 244{
 245	unsigned int i;
 246
 247	if (!t)
 248		return;
 249
 250	while (atomic_read(&t->holders))
 251		msleep(1);
 252	smp_mb();
 253
 254	/* free the indexes */
 255	if (t->depth >= 2)
 256		vfree(t->index[t->depth - 2]);
 257
 258	/* free the targets */
 259	for (i = 0; i < t->num_targets; i++) {
 260		struct dm_target *tgt = t->targets + i;
 261
 262		if (tgt->type->dtr)
 263			tgt->type->dtr(tgt);
 264
 265		dm_put_target_type(tgt->type);
 266	}
 267
 268	vfree(t->highs);
 269
 270	/* free the device list */
 271	free_devices(&t->devices);
 
 272
 273	dm_free_md_mempools(t->mempools);
 274
 275	kfree(t);
 276}
 277
 278void dm_table_get(struct dm_table *t)
 279{
 280	atomic_inc(&t->holders);
 281}
 282EXPORT_SYMBOL(dm_table_get);
 283
 284void dm_table_put(struct dm_table *t)
 285{
 286	if (!t)
 287		return;
 288
 289	smp_mb__before_atomic_dec();
 290	atomic_dec(&t->holders);
 291}
 292EXPORT_SYMBOL(dm_table_put);
 293
 294/*
 295 * Checks to see if we need to extend highs or targets.
 296 */
 297static inline int check_space(struct dm_table *t)
 298{
 299	if (t->num_targets >= t->num_allocated)
 300		return alloc_targets(t, t->num_allocated * 2);
 301
 302	return 0;
 303}
 304
 305/*
 306 * See if we've already got a device in the list.
 307 */
 308static struct dm_dev_internal *find_device(struct list_head *l, dev_t dev)
 309{
 310	struct dm_dev_internal *dd;
 311
 312	list_for_each_entry (dd, l, list)
 313		if (dd->dm_dev.bdev->bd_dev == dev)
 314			return dd;
 315
 316	return NULL;
 317}
 318
 319/*
 320 * Open a device so we can use it as a map destination.
 321 */
 322static int open_dev(struct dm_dev_internal *d, dev_t dev,
 323		    struct mapped_device *md)
 324{
 325	static char *_claim_ptr = "I belong to device-mapper";
 326	struct block_device *bdev;
 327
 328	int r;
 329
 330	BUG_ON(d->dm_dev.bdev);
 331
 332	bdev = blkdev_get_by_dev(dev, d->dm_dev.mode | FMODE_EXCL, _claim_ptr);
 333	if (IS_ERR(bdev))
 334		return PTR_ERR(bdev);
 335
 336	r = bd_link_disk_holder(bdev, dm_disk(md));
 337	if (r) {
 338		blkdev_put(bdev, d->dm_dev.mode | FMODE_EXCL);
 339		return r;
 340	}
 341
 342	d->dm_dev.bdev = bdev;
 343	return 0;
 344}
 345
 346/*
 347 * Close a device that we've been using.
 348 */
 349static void close_dev(struct dm_dev_internal *d, struct mapped_device *md)
 350{
 351	if (!d->dm_dev.bdev)
 352		return;
 353
 354	bd_unlink_disk_holder(d->dm_dev.bdev, dm_disk(md));
 355	blkdev_put(d->dm_dev.bdev, d->dm_dev.mode | FMODE_EXCL);
 356	d->dm_dev.bdev = NULL;
 357}
 358
 359/*
 360 * If possible, this checks an area of a destination device is invalid.
 361 */
 362static int device_area_is_invalid(struct dm_target *ti, struct dm_dev *dev,
 363				  sector_t start, sector_t len, void *data)
 364{
 365	struct request_queue *q;
 366	struct queue_limits *limits = data;
 367	struct block_device *bdev = dev->bdev;
 368	sector_t dev_size =
 369		i_size_read(bdev->bd_inode) >> SECTOR_SHIFT;
 370	unsigned short logical_block_size_sectors =
 371		limits->logical_block_size >> SECTOR_SHIFT;
 372	char b[BDEVNAME_SIZE];
 373
 374	/*
 375	 * Some devices exist without request functions,
 376	 * such as loop devices not yet bound to backing files.
 377	 * Forbid the use of such devices.
 378	 */
 379	q = bdev_get_queue(bdev);
 380	if (!q || !q->make_request_fn) {
 381		DMWARN("%s: %s is not yet initialised: "
 382		       "start=%llu, len=%llu, dev_size=%llu",
 383		       dm_device_name(ti->table->md), bdevname(bdev, b),
 384		       (unsigned long long)start,
 385		       (unsigned long long)len,
 386		       (unsigned long long)dev_size);
 387		return 1;
 388	}
 389
 390	if (!dev_size)
 391		return 0;
 392
 393	if ((start >= dev_size) || (start + len > dev_size)) {
 394		DMWARN("%s: %s too small for target: "
 395		       "start=%llu, len=%llu, dev_size=%llu",
 396		       dm_device_name(ti->table->md), bdevname(bdev, b),
 397		       (unsigned long long)start,
 398		       (unsigned long long)len,
 399		       (unsigned long long)dev_size);
 400		return 1;
 401	}
 402
 403	if (logical_block_size_sectors <= 1)
 404		return 0;
 405
 406	if (start & (logical_block_size_sectors - 1)) {
 407		DMWARN("%s: start=%llu not aligned to h/w "
 408		       "logical block size %u of %s",
 409		       dm_device_name(ti->table->md),
 410		       (unsigned long long)start,
 411		       limits->logical_block_size, bdevname(bdev, b));
 412		return 1;
 413	}
 414
 415	if (len & (logical_block_size_sectors - 1)) {
 416		DMWARN("%s: len=%llu not aligned to h/w "
 417		       "logical block size %u of %s",
 418		       dm_device_name(ti->table->md),
 419		       (unsigned long long)len,
 420		       limits->logical_block_size, bdevname(bdev, b));
 421		return 1;
 422	}
 423
 424	return 0;
 425}
 426
 427/*
 428 * This upgrades the mode on an already open dm_dev, being
 429 * careful to leave things as they were if we fail to reopen the
 430 * device and not to touch the existing bdev field in case
 431 * it is accessed concurrently inside dm_table_any_congested().
 432 */
 433static int upgrade_mode(struct dm_dev_internal *dd, fmode_t new_mode,
 434			struct mapped_device *md)
 435{
 436	int r;
 437	struct dm_dev_internal dd_new, dd_old;
 438
 439	dd_new = dd_old = *dd;
 440
 441	dd_new.dm_dev.mode |= new_mode;
 442	dd_new.dm_dev.bdev = NULL;
 443
 444	r = open_dev(&dd_new, dd->dm_dev.bdev->bd_dev, md);
 445	if (r)
 446		return r;
 447
 448	dd->dm_dev.mode |= new_mode;
 449	close_dev(&dd_old, md);
 450
 451	return 0;
 452}
 453
 454/*
 455 * Add a device to the list, or just increment the usage count if
 456 * it's already present.
 457 */
 458int dm_get_device(struct dm_target *ti, const char *path, fmode_t mode,
 459		  struct dm_dev **result)
 460{
 461	int r;
 462	dev_t uninitialized_var(dev);
 463	struct dm_dev_internal *dd;
 464	unsigned int major, minor;
 465	struct dm_table *t = ti->table;
 466	char dummy;
 467
 468	BUG_ON(!t);
 469
 470	if (sscanf(path, "%u:%u%c", &major, &minor, &dummy) == 2) {
 471		/* Extract the major/minor numbers */
 472		dev = MKDEV(major, minor);
 473		if (MAJOR(dev) != major || MINOR(dev) != minor)
 474			return -EOVERFLOW;
 475	} else {
 476		/* convert the path to a device */
 477		struct block_device *bdev = lookup_bdev(path);
 478
 479		if (IS_ERR(bdev))
 480			return PTR_ERR(bdev);
 481		dev = bdev->bd_dev;
 482		bdput(bdev);
 483	}
 484
 485	dd = find_device(&t->devices, dev);
 486	if (!dd) {
 487		dd = kmalloc(sizeof(*dd), GFP_KERNEL);
 488		if (!dd)
 489			return -ENOMEM;
 490
 491		dd->dm_dev.mode = mode;
 492		dd->dm_dev.bdev = NULL;
 493
 494		if ((r = open_dev(dd, dev, t->md))) {
 495			kfree(dd);
 496			return r;
 497		}
 498
 499		format_dev_t(dd->dm_dev.name, dev);
 500
 501		atomic_set(&dd->count, 0);
 502		list_add(&dd->list, &t->devices);
 503
 504	} else if (dd->dm_dev.mode != (mode | dd->dm_dev.mode)) {
 505		r = upgrade_mode(dd, mode, t->md);
 506		if (r)
 507			return r;
 508	}
 509	atomic_inc(&dd->count);
 510
 511	*result = &dd->dm_dev;
 512	return 0;
 513}
 514EXPORT_SYMBOL(dm_get_device);
 515
 516int dm_set_device_limits(struct dm_target *ti, struct dm_dev *dev,
 517			 sector_t start, sector_t len, void *data)
 518{
 519	struct queue_limits *limits = data;
 520	struct block_device *bdev = dev->bdev;
 521	struct request_queue *q = bdev_get_queue(bdev);
 522	char b[BDEVNAME_SIZE];
 523
 524	if (unlikely(!q)) {
 525		DMWARN("%s: Cannot set limits for nonexistent device %s",
 526		       dm_device_name(ti->table->md), bdevname(bdev, b));
 527		return 0;
 528	}
 529
 530	if (bdev_stack_limits(limits, bdev, start) < 0)
 531		DMWARN("%s: adding target device %s caused an alignment inconsistency: "
 532		       "physical_block_size=%u, logical_block_size=%u, "
 533		       "alignment_offset=%u, start=%llu",
 534		       dm_device_name(ti->table->md), bdevname(bdev, b),
 535		       q->limits.physical_block_size,
 536		       q->limits.logical_block_size,
 537		       q->limits.alignment_offset,
 538		       (unsigned long long) start << SECTOR_SHIFT);
 539
 540	/*
 541	 * Check if merge fn is supported.
 542	 * If not we'll force DM to use PAGE_SIZE or
 543	 * smaller I/O, just to be safe.
 544	 */
 545	if (dm_queue_merge_is_compulsory(q) && !ti->type->merge)
 546		blk_limits_max_hw_sectors(limits,
 547					  (unsigned int) (PAGE_SIZE >> 9));
 548	return 0;
 549}
 550EXPORT_SYMBOL_GPL(dm_set_device_limits);
 551
 552/*
 553 * Decrement a device's use count and remove it if necessary.
 554 */
 555void dm_put_device(struct dm_target *ti, struct dm_dev *d)
 556{
 557	struct dm_dev_internal *dd = container_of(d, struct dm_dev_internal,
 558						  dm_dev);
 559
 560	if (atomic_dec_and_test(&dd->count)) {
 561		close_dev(dd, ti->table->md);
 562		list_del(&dd->list);
 563		kfree(dd);
 564	}
 565}
 566EXPORT_SYMBOL(dm_put_device);
 567
 568/*
 569 * Checks to see if the target joins onto the end of the table.
 570 */
 571static int adjoin(struct dm_table *table, struct dm_target *ti)
 572{
 573	struct dm_target *prev;
 574
 575	if (!table->num_targets)
 576		return !ti->begin;
 577
 578	prev = &table->targets[table->num_targets - 1];
 579	return (ti->begin == (prev->begin + prev->len));
 580}
 581
 582/*
 583 * Used to dynamically allocate the arg array.
 584 */
 585static char **realloc_argv(unsigned *array_size, char **old_argv)
 586{
 587	char **argv;
 588	unsigned new_size;
 589
 590	new_size = *array_size ? *array_size * 2 : 64;
 591	argv = kmalloc(new_size * sizeof(*argv), GFP_KERNEL);
 592	if (argv) {
 593		memcpy(argv, old_argv, *array_size * sizeof(*argv));
 594		*array_size = new_size;
 595	}
 596
 597	kfree(old_argv);
 598	return argv;
 599}
 600
 601/*
 602 * Destructively splits up the argument list to pass to ctr.
 603 */
 604int dm_split_args(int *argc, char ***argvp, char *input)
 605{
 606	char *start, *end = input, *out, **argv = NULL;
 607	unsigned array_size = 0;
 608
 609	*argc = 0;
 610
 611	if (!input) {
 612		*argvp = NULL;
 613		return 0;
 614	}
 615
 616	argv = realloc_argv(&array_size, argv);
 617	if (!argv)
 618		return -ENOMEM;
 619
 620	while (1) {
 621		/* Skip whitespace */
 622		start = skip_spaces(end);
 623
 624		if (!*start)
 625			break;	/* success, we hit the end */
 626
 627		/* 'out' is used to remove any back-quotes */
 628		end = out = start;
 629		while (*end) {
 630			/* Everything apart from '\0' can be quoted */
 631			if (*end == '\\' && *(end + 1)) {
 632				*out++ = *(end + 1);
 633				end += 2;
 634				continue;
 635			}
 636
 637			if (isspace(*end))
 638				break;	/* end of token */
 639
 640			*out++ = *end++;
 641		}
 642
 643		/* have we already filled the array ? */
 644		if ((*argc + 1) > array_size) {
 645			argv = realloc_argv(&array_size, argv);
 646			if (!argv)
 647				return -ENOMEM;
 648		}
 649
 650		/* we know this is whitespace */
 651		if (*end)
 652			end++;
 653
 654		/* terminate the string and put it in the array */
 655		*out = '\0';
 656		argv[*argc] = start;
 657		(*argc)++;
 658	}
 659
 660	*argvp = argv;
 661	return 0;
 662}
 663
 664/*
 665 * Impose necessary and sufficient conditions on a devices's table such
 666 * that any incoming bio which respects its logical_block_size can be
 667 * processed successfully.  If it falls across the boundary between
 668 * two or more targets, the size of each piece it gets split into must
 669 * be compatible with the logical_block_size of the target processing it.
 670 */
 671static int validate_hardware_logical_block_alignment(struct dm_table *table,
 672						 struct queue_limits *limits)
 673{
 674	/*
 675	 * This function uses arithmetic modulo the logical_block_size
 676	 * (in units of 512-byte sectors).
 677	 */
 678	unsigned short device_logical_block_size_sects =
 679		limits->logical_block_size >> SECTOR_SHIFT;
 680
 681	/*
 682	 * Offset of the start of the next table entry, mod logical_block_size.
 683	 */
 684	unsigned short next_target_start = 0;
 685
 686	/*
 687	 * Given an aligned bio that extends beyond the end of a
 688	 * target, how many sectors must the next target handle?
 689	 */
 690	unsigned short remaining = 0;
 691
 692	struct dm_target *uninitialized_var(ti);
 693	struct queue_limits ti_limits;
 694	unsigned i = 0;
 695
 696	/*
 697	 * Check each entry in the table in turn.
 698	 */
 699	while (i < dm_table_get_num_targets(table)) {
 700		ti = dm_table_get_target(table, i++);
 701
 702		blk_set_stacking_limits(&ti_limits);
 703
 704		/* combine all target devices' limits */
 705		if (ti->type->iterate_devices)
 706			ti->type->iterate_devices(ti, dm_set_device_limits,
 707						  &ti_limits);
 708
 709		/*
 710		 * If the remaining sectors fall entirely within this
 711		 * table entry are they compatible with its logical_block_size?
 712		 */
 713		if (remaining < ti->len &&
 714		    remaining & ((ti_limits.logical_block_size >>
 715				  SECTOR_SHIFT) - 1))
 716			break;	/* Error */
 717
 718		next_target_start =
 719		    (unsigned short) ((next_target_start + ti->len) &
 720				      (device_logical_block_size_sects - 1));
 721		remaining = next_target_start ?
 722		    device_logical_block_size_sects - next_target_start : 0;
 723	}
 724
 725	if (remaining) {
 726		DMWARN("%s: table line %u (start sect %llu len %llu) "
 727		       "not aligned to h/w logical block size %u",
 728		       dm_device_name(table->md), i,
 729		       (unsigned long long) ti->begin,
 730		       (unsigned long long) ti->len,
 731		       limits->logical_block_size);
 732		return -EINVAL;
 733	}
 734
 735	return 0;
 736}
 737
 738int dm_table_add_target(struct dm_table *t, const char *type,
 739			sector_t start, sector_t len, char *params)
 740{
 741	int r = -EINVAL, argc;
 742	char **argv;
 743	struct dm_target *tgt;
 744
 745	if (t->singleton) {
 746		DMERR("%s: target type %s must appear alone in table",
 747		      dm_device_name(t->md), t->targets->type->name);
 748		return -EINVAL;
 749	}
 750
 751	if ((r = check_space(t)))
 752		return r;
 753
 754	tgt = t->targets + t->num_targets;
 755	memset(tgt, 0, sizeof(*tgt));
 756
 757	if (!len) {
 758		DMERR("%s: zero-length target", dm_device_name(t->md));
 759		return -EINVAL;
 760	}
 761
 762	tgt->type = dm_get_target_type(type);
 763	if (!tgt->type) {
 764		DMERR("%s: %s: unknown target type", dm_device_name(t->md),
 765		      type);
 766		return -EINVAL;
 767	}
 768
 769	if (dm_target_needs_singleton(tgt->type)) {
 770		if (t->num_targets) {
 771			DMERR("%s: target type %s must appear alone in table",
 772			      dm_device_name(t->md), type);
 773			return -EINVAL;
 774		}
 775		t->singleton = 1;
 776	}
 777
 778	if (dm_target_always_writeable(tgt->type) && !(t->mode & FMODE_WRITE)) {
 779		DMERR("%s: target type %s may not be included in read-only tables",
 780		      dm_device_name(t->md), type);
 781		return -EINVAL;
 782	}
 783
 784	if (t->immutable_target_type) {
 785		if (t->immutable_target_type != tgt->type) {
 786			DMERR("%s: immutable target type %s cannot be mixed with other target types",
 787			      dm_device_name(t->md), t->immutable_target_type->name);
 788			return -EINVAL;
 789		}
 790	} else if (dm_target_is_immutable(tgt->type)) {
 791		if (t->num_targets) {
 792			DMERR("%s: immutable target type %s cannot be mixed with other target types",
 793			      dm_device_name(t->md), tgt->type->name);
 794			return -EINVAL;
 795		}
 796		t->immutable_target_type = tgt->type;
 797	}
 798
 799	tgt->table = t;
 800	tgt->begin = start;
 801	tgt->len = len;
 802	tgt->error = "Unknown error";
 803
 804	/*
 805	 * Does this target adjoin the previous one ?
 806	 */
 807	if (!adjoin(t, tgt)) {
 808		tgt->error = "Gap in table";
 809		r = -EINVAL;
 810		goto bad;
 811	}
 812
 813	r = dm_split_args(&argc, &argv, params);
 814	if (r) {
 815		tgt->error = "couldn't split parameters (insufficient memory)";
 816		goto bad;
 817	}
 818
 819	r = tgt->type->ctr(tgt, argc, argv);
 820	kfree(argv);
 821	if (r)
 822		goto bad;
 823
 824	t->highs[t->num_targets++] = tgt->begin + tgt->len - 1;
 825
 826	if (!tgt->num_discard_requests && tgt->discards_supported)
 827		DMWARN("%s: %s: ignoring discards_supported because num_discard_requests is zero.",
 828		       dm_device_name(t->md), type);
 829
 830	return 0;
 831
 832 bad:
 833	DMERR("%s: %s: %s", dm_device_name(t->md), type, tgt->error);
 834	dm_put_target_type(tgt->type);
 835	return r;
 836}
 837
 838/*
 839 * Target argument parsing helpers.
 840 */
 841static int validate_next_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 842			     unsigned *value, char **error, unsigned grouped)
 843{
 844	const char *arg_str = dm_shift_arg(arg_set);
 845	char dummy;
 846
 847	if (!arg_str ||
 848	    (sscanf(arg_str, "%u%c", value, &dummy) != 1) ||
 849	    (*value < arg->min) ||
 850	    (*value > arg->max) ||
 851	    (grouped && arg_set->argc < *value)) {
 852		*error = arg->error;
 853		return -EINVAL;
 854	}
 855
 856	return 0;
 857}
 858
 859int dm_read_arg(struct dm_arg *arg, struct dm_arg_set *arg_set,
 860		unsigned *value, char **error)
 861{
 862	return validate_next_arg(arg, arg_set, value, error, 0);
 863}
 864EXPORT_SYMBOL(dm_read_arg);
 865
 866int dm_read_arg_group(struct dm_arg *arg, struct dm_arg_set *arg_set,
 867		      unsigned *value, char **error)
 868{
 869	return validate_next_arg(arg, arg_set, value, error, 1);
 870}
 871EXPORT_SYMBOL(dm_read_arg_group);
 872
 873const char *dm_shift_arg(struct dm_arg_set *as)
 874{
 875	char *r;
 876
 877	if (as->argc) {
 878		as->argc--;
 879		r = *as->argv;
 880		as->argv++;
 881		return r;
 882	}
 883
 884	return NULL;
 885}
 886EXPORT_SYMBOL(dm_shift_arg);
 887
 888void dm_consume_args(struct dm_arg_set *as, unsigned num_args)
 889{
 890	BUG_ON(as->argc < num_args);
 891	as->argc -= num_args;
 892	as->argv += num_args;
 893}
 894EXPORT_SYMBOL(dm_consume_args);
 895
 896static int dm_table_set_type(struct dm_table *t)
 897{
 898	unsigned i;
 899	unsigned bio_based = 0, request_based = 0;
 900	struct dm_target *tgt;
 901	struct dm_dev_internal *dd;
 902	struct list_head *devices;
 903
 904	for (i = 0; i < t->num_targets; i++) {
 905		tgt = t->targets + i;
 906		if (dm_target_request_based(tgt))
 907			request_based = 1;
 908		else
 909			bio_based = 1;
 910
 911		if (bio_based && request_based) {
 912			DMWARN("Inconsistent table: different target types"
 913			       " can't be mixed up");
 914			return -EINVAL;
 915		}
 916	}
 917
 918	if (bio_based) {
 919		/* We must use this table as bio-based */
 920		t->type = DM_TYPE_BIO_BASED;
 921		return 0;
 922	}
 923
 924	BUG_ON(!request_based); /* No targets in this table */
 925
 926	/* Non-request-stackable devices can't be used for request-based dm */
 927	devices = dm_table_get_devices(t);
 928	list_for_each_entry(dd, devices, list) {
 929		if (!blk_queue_stackable(bdev_get_queue(dd->dm_dev.bdev))) {
 930			DMWARN("table load rejected: including"
 931			       " non-request-stackable devices");
 932			return -EINVAL;
 933		}
 934	}
 935
 936	/*
 937	 * Request-based dm supports only tables that have a single target now.
 938	 * To support multiple targets, request splitting support is needed,
 939	 * and that needs lots of changes in the block-layer.
 940	 * (e.g. request completion process for partial completion.)
 941	 */
 942	if (t->num_targets > 1) {
 943		DMWARN("Request-based dm doesn't support multiple targets yet");
 944		return -EINVAL;
 945	}
 946
 947	t->type = DM_TYPE_REQUEST_BASED;
 948
 949	return 0;
 950}
 951
 952unsigned dm_table_get_type(struct dm_table *t)
 953{
 954	return t->type;
 955}
 956
 957struct target_type *dm_table_get_immutable_target_type(struct dm_table *t)
 958{
 959	return t->immutable_target_type;
 960}
 961
 962bool dm_table_request_based(struct dm_table *t)
 963{
 964	return dm_table_get_type(t) == DM_TYPE_REQUEST_BASED;
 965}
 966
 967int dm_table_alloc_md_mempools(struct dm_table *t)
 968{
 969	unsigned type = dm_table_get_type(t);
 970
 971	if (unlikely(type == DM_TYPE_NONE)) {
 972		DMWARN("no table type is set, can't allocate mempools");
 973		return -EINVAL;
 974	}
 975
 976	t->mempools = dm_alloc_md_mempools(type, t->integrity_supported);
 977	if (!t->mempools)
 978		return -ENOMEM;
 979
 980	return 0;
 981}
 982
 983void dm_table_free_md_mempools(struct dm_table *t)
 984{
 985	dm_free_md_mempools(t->mempools);
 986	t->mempools = NULL;
 987}
 988
 989struct dm_md_mempools *dm_table_get_md_mempools(struct dm_table *t)
 990{
 991	return t->mempools;
 992}
 993
 994static int setup_indexes(struct dm_table *t)
 995{
 996	int i;
 997	unsigned int total = 0;
 998	sector_t *indexes;
 999
1000	/* allocate the space for *all* the indexes */
1001	for (i = t->depth - 2; i >= 0; i--) {
1002		t->counts[i] = dm_div_up(t->counts[i + 1], CHILDREN_PER_NODE);
1003		total += t->counts[i];
1004	}
1005
1006	indexes = (sector_t *) dm_vcalloc(total, (unsigned long) NODE_SIZE);
1007	if (!indexes)
1008		return -ENOMEM;
1009
1010	/* set up internal nodes, bottom-up */
1011	for (i = t->depth - 2; i >= 0; i--) {
1012		t->index[i] = indexes;
1013		indexes += (KEYS_PER_NODE * t->counts[i]);
1014		setup_btree_index(i, t);
1015	}
1016
1017	return 0;
1018}
1019
1020/*
1021 * Builds the btree to index the map.
1022 */
1023static int dm_table_build_index(struct dm_table *t)
1024{
1025	int r = 0;
1026	unsigned int leaf_nodes;
1027
1028	/* how many indexes will the btree have ? */
1029	leaf_nodes = dm_div_up(t->num_targets, KEYS_PER_NODE);
1030	t->depth = 1 + int_log(leaf_nodes, CHILDREN_PER_NODE);
1031
1032	/* leaf layer has already been set up */
1033	t->counts[t->depth - 1] = leaf_nodes;
1034	t->index[t->depth - 1] = t->highs;
1035
1036	if (t->depth >= 2)
1037		r = setup_indexes(t);
1038
1039	return r;
1040}
1041
1042/*
1043 * Get a disk whose integrity profile reflects the table's profile.
1044 * If %match_all is true, all devices' profiles must match.
1045 * If %match_all is false, all devices must at least have an
1046 * allocated integrity profile; but uninitialized is ok.
1047 * Returns NULL if integrity support was inconsistent or unavailable.
1048 */
1049static struct gendisk * dm_table_get_integrity_disk(struct dm_table *t,
1050						    bool match_all)
1051{
1052	struct list_head *devices = dm_table_get_devices(t);
1053	struct dm_dev_internal *dd = NULL;
1054	struct gendisk *prev_disk = NULL, *template_disk = NULL;
1055
1056	list_for_each_entry(dd, devices, list) {
1057		template_disk = dd->dm_dev.bdev->bd_disk;
1058		if (!blk_get_integrity(template_disk))
1059			goto no_integrity;
1060		if (!match_all && !blk_integrity_is_initialized(template_disk))
1061			continue; /* skip uninitialized profiles */
1062		else if (prev_disk &&
1063			 blk_integrity_compare(prev_disk, template_disk) < 0)
1064			goto no_integrity;
1065		prev_disk = template_disk;
1066	}
1067
1068	return template_disk;
1069
1070no_integrity:
1071	if (prev_disk)
1072		DMWARN("%s: integrity not set: %s and %s profile mismatch",
1073		       dm_device_name(t->md),
1074		       prev_disk->disk_name,
1075		       template_disk->disk_name);
1076	return NULL;
1077}
1078
1079/*
1080 * Register the mapped device for blk_integrity support if
1081 * the underlying devices have an integrity profile.  But all devices
1082 * may not have matching profiles (checking all devices isn't reliable
1083 * during table load because this table may use other DM device(s) which
1084 * must be resumed before they will have an initialized integity profile).
1085 * Stacked DM devices force a 2 stage integrity profile validation:
1086 * 1 - during load, validate all initialized integrity profiles match
1087 * 2 - during resume, validate all integrity profiles match
1088 */
1089static int dm_table_prealloc_integrity(struct dm_table *t, struct mapped_device *md)
1090{
1091	struct gendisk *template_disk = NULL;
1092
1093	template_disk = dm_table_get_integrity_disk(t, false);
1094	if (!template_disk)
1095		return 0;
1096
1097	if (!blk_integrity_is_initialized(dm_disk(md))) {
1098		t->integrity_supported = 1;
1099		return blk_integrity_register(dm_disk(md), NULL);
1100	}
1101
1102	/*
1103	 * If DM device already has an initalized integrity
1104	 * profile the new profile should not conflict.
1105	 */
1106	if (blk_integrity_is_initialized(template_disk) &&
1107	    blk_integrity_compare(dm_disk(md), template_disk) < 0) {
1108		DMWARN("%s: conflict with existing integrity profile: "
1109		       "%s profile mismatch",
1110		       dm_device_name(t->md),
1111		       template_disk->disk_name);
1112		return 1;
1113	}
1114
1115	/* Preserve existing initialized integrity profile */
1116	t->integrity_supported = 1;
1117	return 0;
1118}
1119
1120/*
1121 * Prepares the table for use by building the indices,
1122 * setting the type, and allocating mempools.
1123 */
1124int dm_table_complete(struct dm_table *t)
1125{
1126	int r;
1127
1128	r = dm_table_set_type(t);
1129	if (r) {
1130		DMERR("unable to set table type");
1131		return r;
1132	}
1133
1134	r = dm_table_build_index(t);
1135	if (r) {
1136		DMERR("unable to build btrees");
1137		return r;
1138	}
1139
1140	r = dm_table_prealloc_integrity(t, t->md);
1141	if (r) {
1142		DMERR("could not register integrity profile.");
1143		return r;
1144	}
1145
1146	r = dm_table_alloc_md_mempools(t);
1147	if (r)
1148		DMERR("unable to allocate mempools");
1149
1150	return r;
1151}
1152
1153static DEFINE_MUTEX(_event_lock);
1154void dm_table_event_callback(struct dm_table *t,
1155			     void (*fn)(void *), void *context)
1156{
1157	mutex_lock(&_event_lock);
1158	t->event_fn = fn;
1159	t->event_context = context;
1160	mutex_unlock(&_event_lock);
1161}
1162
1163void dm_table_event(struct dm_table *t)
1164{
1165	/*
1166	 * You can no longer call dm_table_event() from interrupt
1167	 * context, use a bottom half instead.
1168	 */
1169	BUG_ON(in_interrupt());
1170
1171	mutex_lock(&_event_lock);
1172	if (t->event_fn)
1173		t->event_fn(t->event_context);
1174	mutex_unlock(&_event_lock);
1175}
1176EXPORT_SYMBOL(dm_table_event);
1177
1178sector_t dm_table_get_size(struct dm_table *t)
1179{
1180	return t->num_targets ? (t->highs[t->num_targets - 1] + 1) : 0;
1181}
1182EXPORT_SYMBOL(dm_table_get_size);
1183
1184struct dm_target *dm_table_get_target(struct dm_table *t, unsigned int index)
1185{
1186	if (index >= t->num_targets)
1187		return NULL;
1188
1189	return t->targets + index;
1190}
1191
1192/*
1193 * Search the btree for the correct target.
1194 *
1195 * Caller should check returned pointer with dm_target_is_valid()
1196 * to trap I/O beyond end of device.
1197 */
1198struct dm_target *dm_table_find_target(struct dm_table *t, sector_t sector)
1199{
1200	unsigned int l, n = 0, k = 0;
1201	sector_t *node;
1202
1203	for (l = 0; l < t->depth; l++) {
1204		n = get_child(n, k);
1205		node = get_node(t, l, n);
1206
1207		for (k = 0; k < KEYS_PER_NODE; k++)
1208			if (node[k] >= sector)
1209				break;
1210	}
1211
1212	return &t->targets[(KEYS_PER_NODE * n) + k];
1213}
1214
1215/*
1216 * Establish the new table's queue_limits and validate them.
1217 */
1218int dm_calculate_queue_limits(struct dm_table *table,
1219			      struct queue_limits *limits)
1220{
1221	struct dm_target *uninitialized_var(ti);
1222	struct queue_limits ti_limits;
1223	unsigned i = 0;
1224
1225	blk_set_stacking_limits(limits);
1226
1227	while (i < dm_table_get_num_targets(table)) {
1228		blk_set_stacking_limits(&ti_limits);
1229
1230		ti = dm_table_get_target(table, i++);
1231
1232		if (!ti->type->iterate_devices)
1233			goto combine_limits;
1234
1235		/*
1236		 * Combine queue limits of all the devices this target uses.
1237		 */
1238		ti->type->iterate_devices(ti, dm_set_device_limits,
1239					  &ti_limits);
1240
1241		/* Set I/O hints portion of queue limits */
1242		if (ti->type->io_hints)
1243			ti->type->io_hints(ti, &ti_limits);
1244
1245		/*
1246		 * Check each device area is consistent with the target's
1247		 * overall queue limits.
1248		 */
1249		if (ti->type->iterate_devices(ti, device_area_is_invalid,
1250					      &ti_limits))
1251			return -EINVAL;
1252
1253combine_limits:
1254		/*
1255		 * Merge this target's queue limits into the overall limits
1256		 * for the table.
1257		 */
1258		if (blk_stack_limits(limits, &ti_limits, 0) < 0)
1259			DMWARN("%s: adding target device "
1260			       "(start sect %llu len %llu) "
1261			       "caused an alignment inconsistency",
1262			       dm_device_name(table->md),
1263			       (unsigned long long) ti->begin,
1264			       (unsigned long long) ti->len);
1265	}
1266
1267	return validate_hardware_logical_block_alignment(table, limits);
1268}
1269
1270/*
1271 * Set the integrity profile for this device if all devices used have
1272 * matching profiles.  We're quite deep in the resume path but still
1273 * don't know if all devices (particularly DM devices this device
1274 * may be stacked on) have matching profiles.  Even if the profiles
1275 * don't match we have no way to fail (to resume) at this point.
1276 */
1277static void dm_table_set_integrity(struct dm_table *t)
1278{
1279	struct gendisk *template_disk = NULL;
1280
1281	if (!blk_get_integrity(dm_disk(t->md)))
1282		return;
1283
1284	template_disk = dm_table_get_integrity_disk(t, true);
1285	if (template_disk)
1286		blk_integrity_register(dm_disk(t->md),
1287				       blk_get_integrity(template_disk));
1288	else if (blk_integrity_is_initialized(dm_disk(t->md)))
1289		DMWARN("%s: device no longer has a valid integrity profile",
1290		       dm_device_name(t->md));
1291	else
1292		DMWARN("%s: unable to establish an integrity profile",
1293		       dm_device_name(t->md));
1294}
1295
1296static int device_flush_capable(struct dm_target *ti, struct dm_dev *dev,
1297				sector_t start, sector_t len, void *data)
1298{
1299	unsigned flush = (*(unsigned *)data);
1300	struct request_queue *q = bdev_get_queue(dev->bdev);
1301
1302	return q && (q->flush_flags & flush);
1303}
1304
1305static bool dm_table_supports_flush(struct dm_table *t, unsigned flush)
1306{
1307	struct dm_target *ti;
1308	unsigned i = 0;
1309
1310	/*
1311	 * Require at least one underlying device to support flushes.
1312	 * t->devices includes internal dm devices such as mirror logs
1313	 * so we need to use iterate_devices here, which targets
1314	 * supporting flushes must provide.
1315	 */
1316	while (i < dm_table_get_num_targets(t)) {
1317		ti = dm_table_get_target(t, i++);
1318
1319		if (!ti->num_flush_requests)
1320			continue;
1321
1322		if (ti->type->iterate_devices &&
1323		    ti->type->iterate_devices(ti, device_flush_capable, &flush))
1324			return 1;
1325	}
1326
1327	return 0;
1328}
1329
1330static bool dm_table_discard_zeroes_data(struct dm_table *t)
1331{
1332	struct dm_target *ti;
1333	unsigned i = 0;
1334
1335	/* Ensure that all targets supports discard_zeroes_data. */
1336	while (i < dm_table_get_num_targets(t)) {
1337		ti = dm_table_get_target(t, i++);
1338
1339		if (ti->discard_zeroes_data_unsupported)
1340			return 0;
1341	}
1342
1343	return 1;
1344}
1345
1346static int device_is_nonrot(struct dm_target *ti, struct dm_dev *dev,
1347			    sector_t start, sector_t len, void *data)
1348{
1349	struct request_queue *q = bdev_get_queue(dev->bdev);
1350
1351	return q && blk_queue_nonrot(q);
1352}
1353
1354static int device_is_not_random(struct dm_target *ti, struct dm_dev *dev,
1355			     sector_t start, sector_t len, void *data)
1356{
1357	struct request_queue *q = bdev_get_queue(dev->bdev);
1358
1359	return q && !blk_queue_add_random(q);
1360}
1361
1362static bool dm_table_all_devices_attribute(struct dm_table *t,
1363					   iterate_devices_callout_fn func)
1364{
1365	struct dm_target *ti;
1366	unsigned i = 0;
1367
1368	while (i < dm_table_get_num_targets(t)) {
1369		ti = dm_table_get_target(t, i++);
1370
1371		if (!ti->type->iterate_devices ||
1372		    !ti->type->iterate_devices(ti, func, NULL))
1373			return 0;
1374	}
1375
1376	return 1;
1377}
1378
1379void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q,
1380			       struct queue_limits *limits)
1381{
1382	unsigned flush = 0;
1383
1384	/*
1385	 * Copy table's limits to the DM device's request_queue
1386	 */
1387	q->limits = *limits;
1388
1389	if (!dm_table_supports_discards(t))
1390		queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
1391	else
1392		queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
1393
1394	if (dm_table_supports_flush(t, REQ_FLUSH)) {
1395		flush |= REQ_FLUSH;
1396		if (dm_table_supports_flush(t, REQ_FUA))
1397			flush |= REQ_FUA;
1398	}
1399	blk_queue_flush(q, flush);
1400
1401	if (!dm_table_discard_zeroes_data(t))
1402		q->limits.discard_zeroes_data = 0;
1403
1404	/* Ensure that all underlying devices are non-rotational. */
1405	if (dm_table_all_devices_attribute(t, device_is_nonrot))
1406		queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
1407	else
1408		queue_flag_clear_unlocked(QUEUE_FLAG_NONROT, q);
1409
1410	dm_table_set_integrity(t);
1411
1412	/*
1413	 * Determine whether or not this queue's I/O timings contribute
1414	 * to the entropy pool, Only request-based targets use this.
1415	 * Clear QUEUE_FLAG_ADD_RANDOM if any underlying device does not
1416	 * have it set.
1417	 */
1418	if (blk_queue_add_random(q) && dm_table_all_devices_attribute(t, device_is_not_random))
1419		queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
1420
1421	/*
1422	 * QUEUE_FLAG_STACKABLE must be set after all queue settings are
1423	 * visible to other CPUs because, once the flag is set, incoming bios
1424	 * are processed by request-based dm, which refers to the queue
1425	 * settings.
1426	 * Until the flag set, bios are passed to bio-based dm and queued to
1427	 * md->deferred where queue settings are not needed yet.
1428	 * Those bios are passed to request-based dm at the resume time.
1429	 */
1430	smp_mb();
1431	if (dm_table_request_based(t))
1432		queue_flag_set_unlocked(QUEUE_FLAG_STACKABLE, q);
1433}
1434
1435unsigned int dm_table_get_num_targets(struct dm_table *t)
1436{
1437	return t->num_targets;
1438}
1439
1440struct list_head *dm_table_get_devices(struct dm_table *t)
1441{
1442	return &t->devices;
1443}
1444
1445fmode_t dm_table_get_mode(struct dm_table *t)
1446{
1447	return t->mode;
1448}
1449EXPORT_SYMBOL(dm_table_get_mode);
1450
1451static void suspend_targets(struct dm_table *t, unsigned postsuspend)
1452{
1453	int i = t->num_targets;
1454	struct dm_target *ti = t->targets;
1455
1456	while (i--) {
1457		if (postsuspend) {
1458			if (ti->type->postsuspend)
1459				ti->type->postsuspend(ti);
1460		} else if (ti->type->presuspend)
1461			ti->type->presuspend(ti);
1462
1463		ti++;
1464	}
1465}
1466
1467void dm_table_presuspend_targets(struct dm_table *t)
1468{
1469	if (!t)
1470		return;
1471
1472	suspend_targets(t, 0);
1473}
1474
1475void dm_table_postsuspend_targets(struct dm_table *t)
1476{
1477	if (!t)
1478		return;
1479
1480	suspend_targets(t, 1);
1481}
1482
1483int dm_table_resume_targets(struct dm_table *t)
1484{
1485	int i, r = 0;
1486
1487	for (i = 0; i < t->num_targets; i++) {
1488		struct dm_target *ti = t->targets + i;
1489
1490		if (!ti->type->preresume)
1491			continue;
1492
1493		r = ti->type->preresume(ti);
1494		if (r)
1495			return r;
1496	}
1497
1498	for (i = 0; i < t->num_targets; i++) {
1499		struct dm_target *ti = t->targets + i;
1500
1501		if (ti->type->resume)
1502			ti->type->resume(ti);
1503	}
1504
1505	return 0;
1506}
1507
1508void dm_table_add_target_callbacks(struct dm_table *t, struct dm_target_callbacks *cb)
1509{
1510	list_add(&cb->list, &t->target_callbacks);
1511}
1512EXPORT_SYMBOL_GPL(dm_table_add_target_callbacks);
1513
1514int dm_table_any_congested(struct dm_table *t, int bdi_bits)
1515{
1516	struct dm_dev_internal *dd;
1517	struct list_head *devices = dm_table_get_devices(t);
1518	struct dm_target_callbacks *cb;
1519	int r = 0;
1520
1521	list_for_each_entry(dd, devices, list) {
1522		struct request_queue *q = bdev_get_queue(dd->dm_dev.bdev);
1523		char b[BDEVNAME_SIZE];
1524
1525		if (likely(q))
1526			r |= bdi_congested(&q->backing_dev_info, bdi_bits);
1527		else
1528			DMWARN_LIMIT("%s: any_congested: nonexistent device %s",
1529				     dm_device_name(t->md),
1530				     bdevname(dd->dm_dev.bdev, b));
1531	}
1532
1533	list_for_each_entry(cb, &t->target_callbacks, list)
1534		if (cb->congested_fn)
1535			r |= cb->congested_fn(cb, bdi_bits);
1536
1537	return r;
1538}
1539
1540int dm_table_any_busy_target(struct dm_table *t)
1541{
1542	unsigned i;
1543	struct dm_target *ti;
1544
1545	for (i = 0; i < t->num_targets; i++) {
1546		ti = t->targets + i;
1547		if (ti->type->busy && ti->type->busy(ti))
1548			return 1;
1549	}
1550
1551	return 0;
1552}
1553
1554struct mapped_device *dm_table_get_md(struct dm_table *t)
1555{
1556	return t->md;
1557}
1558EXPORT_SYMBOL(dm_table_get_md);
1559
1560static int device_discard_capable(struct dm_target *ti, struct dm_dev *dev,
1561				  sector_t start, sector_t len, void *data)
1562{
1563	struct request_queue *q = bdev_get_queue(dev->bdev);
1564
1565	return q && blk_queue_discard(q);
1566}
1567
1568bool dm_table_supports_discards(struct dm_table *t)
1569{
1570	struct dm_target *ti;
1571	unsigned i = 0;
1572
1573	/*
1574	 * Unless any target used by the table set discards_supported,
1575	 * require at least one underlying device to support discards.
1576	 * t->devices includes internal dm devices such as mirror logs
1577	 * so we need to use iterate_devices here, which targets
1578	 * supporting discard selectively must provide.
1579	 */
1580	while (i < dm_table_get_num_targets(t)) {
1581		ti = dm_table_get_target(t, i++);
1582
1583		if (!ti->num_discard_requests)
1584			continue;
1585
1586		if (ti->discards_supported)
1587			return 1;
1588
1589		if (ti->type->iterate_devices &&
1590		    ti->type->iterate_devices(ti, device_discard_capable, NULL))
1591			return 1;
1592	}
1593
1594	return 0;
1595}